nvidia 980 8gb release date

davegl1234

Weaksauce
Joined
Aug 14, 2010
Messages
75
Hi,

I was wondering if anyone has any idea when the 8gb of the 980 will be launched?

I'm tempted to wait, if its probably going to be within the next couple of months.

Would you expect EVGA to offer a step-up to this, from the 4gb versions?

Thanks
 
Either we'll see 8GB 980's or we'll see a 980Ti or Titan 2 or some such with 6/8GB.
 
and hopefully it won't cost a thousand dollars. watch, it will sell for $999.99.
 
if Nvidia is smart they will only put 8GB on the Titan 2...will sell a ton of over-priced Titan's and will give them more perceived value
 
Can 8GB VRAM even be leveraged with a 256 bit bus?

Seems the cards are going to run out of horsepower and bandwidth long before they encounter RAM issues.
 
If we look at current trends it looks like the upcoming games will have a much larger increase in memory requirements than computational requirements.

This makes sense when consider that the largest improvement by far with this console generation was local (for the System/GPU) memory capacity (even just the usable portion for games). The increase was much higher than the increase in both CPU and GPU computational power, and even more so than storage system performance and capacity.

You then need to consider for SLI in which you effectively add computational power but memory capacity remains static.

Can 8GB VRAM even be leveraged with a 256 bit bus?

To address this directly from a hardware perspective yes and no. Right now the highest density GDDR5 is 4Gb (gigabits) chips I believe. This means with a 8x32 config (256 bit bus) you top out at 4GB (gigabyte) of memory. To get 8GB you'd need to run at 8+8/16 clamshell mode (essentially 1 chip per side). 8GB would be the maximum limit for a current 256bit card using GDDR5 however. This is actually how the PS4 is configured as well.
 
Last edited:
You then need to consider for SLI in which you effectively add computational power but memory capacity remains static.
That's where the bandwidth problem is going to show.

A single 980 doesn't have the grunt to effectively use 8GB VRAM and multiple 980's are going to choke on the 256 bus.
 
I've seen Gibbo's posts (over at OCUK), he said he didn't even hear about the plans to make those let alone something being worked on right now, so i wouldn't hold my breath. It is weird though that manufacturers are silent on that matter :confused:

As far as step up within the same series, that's was always a no-no with 780 6GB being the only exception i know of, so i wouldn't count on that either.
 
That's where the bandwidth problem is going to show.

A single 980 doesn't have the grunt to effectively use 8GB VRAM and multiple 980's are going to choke on the 256 bus.

Yes and no. It's more workload dependent than hardware dependent.

If you want to look at a generalization for game graphics the data that requires very high bandwidth is somewhat inverse to the size (for modern/current workloads, and seemingly trends going forward). So for example, on one extreme (but common usage), if you want to populate that VRAM simply by much larger textures the bandwidth will not be a limitation.

In terms of bandwidth and SLI, while SLI doesn't exactly directly add bandwidth together it does mitigate the issue since you effectively can take longer (simplified) for each card to render. So in practice you can look at it as adding bandwidth together. You can look at SLI scaling at 1080p vs at 4k for instance (resolution increase being a common scaling issue for bandwidth limited cards), if memory bandwidth was the limiting factor (in real usage) you'd see massive diminishing scaling as resolution/AA moves up. But basically the only thing SLI has no impact on at all is actual VRAM capacity.

So in terms of real usage limitations it basically comes down to workload. If you only require higher VRAM usage from things that also require more computational performance, such as resolution, to a point that the 980 (or multiple 980s) cannot cope than the VRAM won't help you. But not everything that requires more VRAM also requires much higher corresponding computational power.
 
Back
Top